Cerebrovascular disease risk assessment method and device based on large model agent

Abstract

The application provides a cerebral vascular disease risk assessment method and device based on a large model intelligent agent, the method comprising: acquiring multi-source information of a user, the multi-source information of the user representing a comprehensive health status of the user; according to the multi-source user information, combining a cerebral vascular disease knowledge base, acquiring corresponding medical guideline knowledge fragments as context reference materials; inputting the multi-source user information and the context reference materials into a health status assessment model to obtain a health status assessment result output by the health status assessment model, the health status assessment result comprising a cerebral vascular disease risk assessment result and health guidance suggestions; through the multi-source user information and the introduction of the cerebral vascular disease knowledge base, the model can simultaneously consider multiple factors such as physiological indicators, medical history backgrounds and lifestyles of the user, and combined with the implied medical logic, the evaluation process is ensured to have a basis, the professional degree and the reliability of the evaluation result are significantly improved, and the personalized generation of the suggestion content is realized.

Description

Technical Field

[0001] This invention relates to the field of artificial intelligence technology, and in particular to a method and apparatus for assessing the risk of cerebrovascular diseases based on a large-scale intelligent model. Background Technology

[0002] Cerebrovascular diseases, such as stroke, have high incidence, high disability rates, and high mortality rates. Their occurrence and development are closely related to multiple risk factors, including hypertension, diabetes, hyperlipidemia, smoking, and unhealthy lifestyles. With the advancement of the Healthy China strategy, primary prevention and pre-disease management of cerebrovascular diseases have become crucial links in reducing the disease burden. Currently, the health management industry is in a transitional period from informatization to intelligentization. Although physical examinations and health screenings are quite widespread, personalized health interventions and guidance still mainly rely on the experience and judgment of professional physicians.

[0003] Currently, the pre-illness management and health guidance for cerebrovascular diseases are mainly conducted in two ways. One is the traditional manual-led model, which relies on hospitals or primary healthcare institutions. Professional physicians conduct face-to-face consultations or follow-ups, combining residents' physical examination reports and past medical history to conduct a comprehensive assessment and provide verbal lifestyle guidance. The other is the digitally assisted model, which deploys physical examination report interpretation systems or health management software. Based on rule engines or threshold judgment technology, and through a preset medical knowledge base, the system statically matches the user's examination indicator values ​​with standard ranges, and then outputs pre-set fixed text conclusions as health advice.

[0004] However, the aforementioned human-led approach is highly dependent on professional physician resources. Limited by the number and energy of doctors, it is difficult to cover the long-term, high-frequency management needs of a large population, and the quality of services varies. The digital-assisted approach lacks deep reasoning ability and can only perform superficial threshold comparison and rule matching. It is difficult to comprehensively analyze the multimodal characteristics, behavioral habits and dynamic changes of indicators of individuals, resulting in assessment results that often lack depth and personalization and are difficult to simulate the complex logical reasoning process of clinical diagnosis and treatment. Summary of the Invention

[0005] This invention provides a method and device for assessing the risk of cerebrovascular diseases based on a large model intelligent agent, which solves the shortcomings of existing technologies that lack deep reasoning ability, resulting in a lack of depth and personalization in assessment results. While improving the accuracy of assessment results, it ensures the personalized generation of recommendations, realizing the integrated and precise output of risk and intervention and a closed loop of personalized health management.

[0006] This invention provides a method for assessing the risk of cerebrovascular diseases based on a large-scale intelligent model, comprising: acquiring multi-source user information, which represents the user's comprehensive health status; acquiring corresponding medical guideline knowledge fragments based on the multi-source user information and a knowledge base in the field of cerebrovascular diseases, as contextual reference material; inputting the multi-source user information and the contextual reference material into a health status assessment model to obtain the health status assessment result output by the health status assessment model, which includes a cerebrovascular disease risk assessment result and health guidance suggestions; wherein, the health status assessment model is trained based on the user's multi-source training data and the corresponding health status assessment labels of the user's multi-source training data.

[0007] According to the present invention, a method for assessing the risk of cerebrovascular diseases based on a large-scale intelligent agent is provided. The method utilizes multi-source user information, including basic user information, past medical history and family history, physical examination and testing indicators, lifestyle information, and subjective health descriptions. Based on this multi-source user information and combined with a knowledge base in the field of cerebrovascular diseases, corresponding medical guideline knowledge fragments are obtained as contextual reference materials. This includes: performing structured encoding and numerical normalization on the basic user information and physical examination and testing indicators to obtain structured feature vectors; and performing natural language processing and key entity extraction on the past medical history and family history, lifestyle information, and subjective health descriptions to obtain textual semantic feature vectors. The structured feature vectors and textual semantic feature vectors are then fused across modally to obtain multimodal fused feature vectors. These multimodal fused feature vectors are then used to retrieve corresponding medical guideline knowledge fragments from a pre-constructed knowledge base in the field of cerebrovascular diseases, serving as contextual reference materials.

[0008] According to the present invention, a method for assessing the risk of cerebrovascular diseases based on a large-scale intelligent model provides health guidance suggestions including lifestyle adjustment suggestions, key intervention prompts, and follow-up suggestions. The method involves inputting multi-source user information and contextual reference materials into a health status assessment model to obtain the health status assessment results output by the model. This includes: inputting multi-source user information and the contextual reference materials into the health status assessment model; analyzing the multi-source user information based on the contextual reference materials to identify the user's risk factors; combining these with corresponding risk weights to generate a cerebrovascular disease risk assessment result containing a list of risk factors and a comprehensive risk level; identifying the abnormal indicator features and key entities with the highest risk contribution based on the multi-source user information to generate key intervention prompts; determining intervention targets based on the risk factor list and matching corresponding intervention measures to generate corresponding lifestyle adjustment suggestions; determining the follow-up period based on the comprehensive risk level; and determining follow-up items based on the abnormal indicator features and key entities to generate follow-up suggestions.

[0009] According to the present invention, a method for assessing the risk of cerebrovascular diseases based on a large model intelligent agent includes, after obtaining the health status assessment result output by the health status assessment model, the following steps are taken: sending the cerebrovascular disease risk assessment result and health guidance suggestions to a target user, and receiving the result returned by the target user; the result includes the content after the target user confirms, modifies, and / or supplements the cerebrovascular disease risk assessment result and health guidance suggestions; analyzing the differences between the cerebrovascular disease risk assessment result and the result returned by the target user based on the cerebrovascular disease risk assessment result and health guidance suggestions, and extracting correction features reflecting professional knowledge logic; converting the correction features into structured knowledge memory rules, and updating the cerebrovascular disease domain knowledge base; the knowledge memory rules are used to constrain the generation strategy during the model reasoning process.

[0010] According to the present invention, a method for assessing the risk of cerebrovascular disease based on a large model intelligent agent analyzes the differences between the cerebrovascular disease risk assessment results and health guidance suggestions and the results returned by the target user, and extracts correction features reflecting professional knowledge logic. The method includes: semantically aligning the cerebrovascular disease risk assessment results and health guidance suggestions with the results returned by the target user to identify modified target text fragments; classifying the target text fragments to determine the modification type, which includes at least one of medical error correction, personalized supplementation, and expression optimization; and invoking the corresponding entity relationship extraction strategy according to the modification type to extract key entities and relationships between entities from the target text fragments, constructing condition-conclusion pairs as correction features reflecting professional knowledge logic.

[0011] According to the present invention, a method for assessing the risk of cerebrovascular diseases based on a large model intelligent agent includes an entity relationship extraction strategy as follows: when the modification type is determined to be medical error correction, key entities containing medical contraindications or error risk factors and their inter-entity relationships are extracted, and conditional conclusion pairs are constructed in conjunction with intervention measures after correction; when the modification type is determined to be personalized supplementation, key entities containing user-personalized attribute characteristics of non-general medical advice and their inter-entity relationships are extracted, and conditional conclusion pairs are constructed in conjunction with newly added specific intervention plans; when the modification type is determined to be expression optimization, key entities containing the initial descriptive context and their inter-entity relationships are extracted, and conditional conclusion pairs are constructed in conjunction with text expression paradigms that conform to medical standards or user understanding habits.

[0012] According to the present invention, a method for assessing the risk of cerebrovascular diseases based on a large model intelligent agent, after receiving the results returned by the target user, further includes: obtaining the user's case based on the results and the corresponding user's multi-source information; extracting and classifying the risk factors in the user's case when the user meets the conditions for stroke risk based on the user's case; updating the classified risk factors and corresponding user characteristics to the cerebrovascular disease domain knowledge base; and structuring the health guidance suggestions in the results, converting them into standardized intervention strategy items, and updating them to the cerebrovascular disease domain knowledge base.

[0013] According to the present invention, a method for assessing the risk of cerebrovascular diseases based on a large model intelligent agent transforms modified features into structured knowledge memory rules, including: semantically abstracting the modified features and extracting the medical entities involved in the modification and the modification intent; associating and matching the medical entities with user multi-source information to locate the user feature conditions that trigger the modification from the user multi-source information; determining the rule storage attributes according to the modification intent, establishing a mapping relationship between the user feature conditions and the modified health guidance suggestions, forming a general guidance rule or preference pattern containing the rule storage attributes, and obtaining the knowledge memory rules.

[0014] According to the present invention, a method for assessing the risk of cerebrovascular disease based on a large-scale intelligent agent includes the following steps before inputting multi-source user information into a health status assessment model: acquiring human-computer collaborative interaction data during the historical health guidance generation process, wherein the human-computer collaborative interaction data includes user multi-source training data and health status assessment labels corresponding to the user multi-source training data, and the status assessment labels are obtained by correcting the initial assessment results and suggestions obtained by the large model based on the user multi-source training data; using the user multi-source training data as input data for training, and using the risk assessment labels and health guidance labels as labels for training, training the large model to obtain a health status assessment model for assessing the risk of cerebrovascular disease and providing health guidance to users.

[0015] This invention also provides a cerebrovascular disease risk assessment device based on a large-scale intelligent model, comprising: an information acquisition module for acquiring multi-source user information, which represents the user's comprehensive health status; a knowledge retrieval module for acquiring corresponding medical guideline knowledge fragments based on the multi-source user information and a cerebrovascular disease domain knowledge base, as contextual reference material; and a risk assessment module for inputting the multi-source user information and contextual reference material into a health status assessment model to obtain the health status assessment result output by the health status assessment model, which includes cerebrovascular disease risk assessment results and health guidance suggestions; wherein, the health status assessment model is trained based on user multi-source training data and the corresponding health status assessment labels of the user multi-source training data.

[0016] The present invention also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the cerebrovascular disease risk assessment method based on large model intelligent agents as described above.

[0017] The present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the cerebrovascular disease risk assessment method based on a large model intelligent agent as described above.

[0018] The present invention also provides a computer program product, including a computer program that, when executed by a processor, implements the cerebrovascular disease risk assessment method based on a large model intelligent agent as described above.

[0019] The cerebrovascular disease risk assessment method and device based on a large-scale intelligent model provided by this invention acquires multi-source information representing the user's comprehensive health status, enabling the model to simultaneously consider multiple factors such as the user's physiological indicators, medical history, and lifestyle. This provides rich data support for generating in-depth assessment results. Furthermore, it incorporates medical knowledge fragments retrieved from a cerebrovascular disease knowledge base to supplement the model with the implicit medical logic within the user's multi-source information, effectively avoiding the illusion problem that models are prone to in the medical field. This ensures that the assessment process is based on evidence, significantly improving the professionalism and credibility of the assessment results. The model not only outputs risk levels but also reveals deep pathological connections between different risk factors, elevating the assessment results from simple numerical predictions to in-depth diagnostic analysis with medical logic. This allows for the direct generation of customized health guidance suggestions based on the user's unique risk characteristics. While improving the accuracy of the assessment results, it ensures that the suggestions are not generic popular science but rather action guidelines tailored to the user's individual risk factor list, achieving integrated and precise output of risk and intervention, and a closed loop of personalized health management. Attached Figure Description

[0020] To more clearly illustrate the technical solutions in this invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0021] Figure 1 This is one of the flowcharts of the cerebrovascular disease risk assessment method based on a large model intelligent agent provided by the present invention; Figure 2 This is the second flowchart of the cerebrovascular disease risk assessment method based on a large model intelligent agent provided by the present invention; Figure 3 This is a schematic diagram of the cerebrovascular disease risk assessment device based on a large model intelligent agent provided by the present invention; Figure 4 This is a schematic diagram of the structure of the electronic device provided by the present invention. Detailed Implementation

[0022] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this invention. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without creative effort are within the scope of protection of this invention.

[0023] Figure 1 This is a flowchart illustrating the cerebrovascular disease risk assessment method based on a large-scale intelligent agent provided by the present invention, as shown below. Figure 1 As shown, the method includes the following: S11, Obtain multi-source information of users, which represents the comprehensive health status of users; S12. Based on multi-source user information and combined with a knowledge base in the field of cerebrovascular diseases, obtain relevant medical guideline knowledge fragments as contextual reference materials; S13, input multi-source user information and contextual reference materials into the health status assessment model to obtain the health status assessment results output by the health status assessment model. The health status assessment results include cerebrovascular disease risk assessment results and health guidance suggestions; wherein, the health status assessment model is trained based on user multi-source training data and the health status assessment labels corresponding to the user multi-source training data.

[0024] It should be noted that the following will be combined with specific details. Figure 2 This invention describes a method for assessing the risk of cerebrovascular diseases based on a large model intelligent agent.

[0025] Step S11: Obtain multi-source information of the user, which represents the user's comprehensive health status.

[0026] It should be added that the multi-source user information includes basic user information, past medical history and family history, physical examination and test results, lifestyle information, and subjective health descriptions. Basic user information includes age and gender, physical examination and test results include blood pressure, blood lipids, and blood sugar, and lifestyle information includes smoking, alcohol consumption, exercise, and dietary habits. Furthermore, after obtaining the multi-source user information, the data from different sources is structured and timestamped to support dynamic analysis of the user's health status.

[0027] Step S12: Based on multi-source user information and combined with the knowledge base in the field of cerebrovascular diseases, obtain the corresponding medical guideline knowledge fragments as contextual reference materials.

[0028] It should be added that the knowledge base in the field of cerebrovascular diseases was first built based on the guidelines and standards for the prevention of cerebrovascular diseases, clinical expert advice, various typical cases, and the risk factors obtained by classifying and decomposing the corresponding typical cases.

[0029] In this embodiment, based on multi-source user information and combined with a knowledge base in the field of cerebrovascular diseases, corresponding medical guideline knowledge fragments are obtained as contextual reference materials. This includes: performing structured encoding and numerical normalization on user basic information and physical examination and examination indicators to obtain structured feature vectors; and performing natural language processing and key entity extraction on past medical history, family history, lifestyle information, and subjective health description information to obtain text semantic feature vectors; fusing the structured feature vectors and text semantic feature vectors across modalities to obtain multimodal fusion feature vectors; and using the multimodal fusion feature vectors to search in a pre-constructed knowledge base in the field of cerebrovascular diseases to obtain corresponding medical guideline knowledge fragments as contextual reference materials.

[0030] It should be noted that for numerical data such as physical examination indicators, normalization is used to eliminate dimensional differences. For textual data such as medical history and subjective descriptions, key entity extraction is used to accurately extract the core medical semantics from unstructured text. This preserves the accuracy of quantitative data while uncovering the deep semantics of qualitative text. Through cross-modal fusion, physical examination values ​​and textual descriptions rich in background information are mapped to the same feature space, thereby constructing a comprehensive user profile that includes physiological state and behavioral patterns. This enhances the information density and discriminative ability of feature expression. Using fused feature vectors containing comprehensive user information as retrieval queries, medical guideline fragments that highly match the current user can be accurately located in the knowledge base. This deep semantic-based retrieval method not only matches explicit keywords but also links to implicit complication risks or lifestyle associations, ensuring that the retrieved reference materials are authoritative and tailored to the user. This provides high-value prior knowledge for subsequent evaluation models, ensuring effective support for the interpretability and professional depth of the evaluation results.

[0031] Step S13: Input multi-source user information and contextual reference materials into the health status assessment model to obtain the health status assessment results output by the health status assessment model. The health status assessment results include cerebrovascular disease risk assessment results and health guidance suggestions. The health status assessment model is trained based on user multi-source training data and the health status assessment labels corresponding to the user multi-source training data.

[0032] Specifically, the health guidance recommendations include lifestyle adjustment suggestions, key intervention tips, and follow-up recommendations. Multi-source user information and contextual reference materials are input into a health status assessment model to obtain the model's output health status assessment results. These include: inputting multi-source user information and contextual reference materials into the health status assessment model to analyze the multi-source user information based on the contextual reference materials, identifying the user's risk factors, and combining them with corresponding risk weights to generate a cerebrovascular disease risk assessment result containing a list of risk factors and a comprehensive risk level; identifying the abnormal indicator features and key entities with the highest risk contribution based on the multi-source user information, generating key intervention tips; determining intervention targets based on the risk factor list, matching corresponding intervention measures, and generating corresponding lifestyle adjustment suggestions; determining the follow-up period based on the comprehensive risk level, and determining follow-up items based on abnormal indicator features and key entities, generating follow-up recommendations.

[0033] It should be noted that by combining contextual reference materials and risk weights to analyze multi-source user information, the assessment process is made evidence-based and quantifiable, and the results are made transparent. This allows users to intuitively understand why they are assessed as high-risk, transforming abstract algorithmic scoring into understandable risk attribution, significantly improving the interpretability and medical credibility of the assessment results. Furthermore, an attention mechanism is introduced, enabling the model to focus on the core indicators or key entities that contribute the most to risk, ensuring that intervention suggestions are targeted and effective. This helps users quickly grasp the main issues amidst complex health data, improving the efficiency of health management. Additionally, intervention goals are derived and matched with measures based on a specific list of risk factors, ensuring that every individual's risk level is accurately assessed. The lifestyle recommendations are clearly targeted, ensuring that the generated health advice is not just generic popular science text, but a practical guide that truly fits the individual pathological characteristics of users. This significantly improves the feasibility of the recommendations. By converting risk levels into specific follow-up cycles and abnormal characteristics into specific re-examination items, a long-term health management timetable is constructed. This effectively fills the gap in the process from assessment to rehabilitation, helping to achieve early detection, early intervention, and full-cycle management of cerebrovascular diseases. It also shortens the early health consultation process, allows some health consultations to be completed at home, reduces the number of people going to the hospital, reduces medical service expenditures, lowers medical costs, and reduces the current medical workload for both doctors and patients.

[0034] In one optional embodiment, before inputting multi-source user information into the health status assessment model, the method includes: acquiring human-computer collaboration interaction data from the historical health guidance generation process, wherein the human-computer collaboration interaction data includes user multi-source training data and health status assessment labels corresponding to the user multi-source training data, and the status assessment labels are obtained by correcting the initial assessment results and suggestions obtained by the large model based on the user multi-source training data; using the user multi-source training data as input data for training, and using the risk assessment labels and health guidance labels as labels for training, the large model is trained to obtain a health status assessment model for assessing the user's cerebrovascular disease risk and providing health guidance.

[0035] It should be added that the state assessment label includes the corresponding corrected initial assessment results and suggestions. The large model can be an existing trained large model or a large model pre-trained using knowledge in the field of cerebrovascular diseases. The specific choice can be made according to actual needs and prior experience, and no further restrictions are made here. Furthermore, by pre-training on a domain knowledge base, the large model has already mastered the core concepts, pathological logic, and association rules of the cerebrovascular disease field before formally facing specific user data. This significantly shortens the convergence time for subsequent tasks and reduces the risk of non-medical logical fallacies. By introducing a human-computer collaboration mechanism, the corrections made by experts to the initial model results are used as labels, cleverly distilling the clinical experience of medical experts into the dataset. This not only corrects potential cognitive errors in the model but also ensures the rigor and authority of the training labels in medicine. This allows the model to learn the implicit knowledge and judgment standards of human experts when handling complex cases. Thus, on the knowledge base built by pre-training, supervised fine-tuning is performed using high-quality human-computer collaborative interaction data. This allows the model to retain the generalization ability of the large model while enhancing its professionalism in risk assessment and health guidance generation tasks through expert-corrected data, resulting in a health status assessment model that combines medical literacy and practical ability.

[0036] In one alternative embodiment, reference Figure 2 After obtaining the health status assessment results output by the health status assessment model, the process includes: sending the cerebrovascular disease risk assessment results and health guidance suggestions to the target user, and receiving the results returned by the target user; the results include the content after the target user confirms, modifies, and / or supplements the cerebrovascular disease risk assessment results and health guidance suggestions; analyzing the differences between the cerebrovascular disease risk assessment results and health guidance suggestions and the results returned by the target user, and extracting correction features that reflect the logic of professional knowledge; transforming the correction features into structured knowledge memory rules, and updating the cerebrovascular disease domain knowledge base; the knowledge memory rules are used to constrain the generation strategy during the model reasoning process.

[0037] It should be noted that the target users are pre-determined professionals or experts with relevant medical knowledge. No further restrictions are imposed here. By continuously updating the knowledge base in the field of cerebrovascular diseases, long-term memory information from the knowledge base is introduced into the process of generating cerebrovascular disease risk assessment results and health guidance using the health status assessment model. This constrains or guides the reasoning process or generation strategy of the artificial intelligence model, enabling the model to prioritize generating health guidance suggestions that are more in line with professional consensus in similar user scenarios. For example, when the system faces users with similar risk factors and concurrent situations again, the model can automatically refer to historical human-computer collaborative memory to avoid repeatedly generating unreasonable suggestions that have been revised multiple times. Furthermore, by introducing a confirmation, modification, and supplementation mechanism for target users, specific users are given the right to review the assessment results. User feedback is also transformed into valuable, high-quality data to ensure timely detection and correction of model-generated biases, guaranteeing the accuracy of the final delivery. This allows the AI ​​model to absorb the corrective experience of professionals, accurately pinpointing weaknesses in the model's reasoning by comparing and analyzing the original recommendations with the modified content from the target users. The model can then extract the corrective features leading to these discrepancies, uncover the underlying medical logic behind the modifications, and provide precise material for knowledge updates. These corrective features are then solidified into knowledge memory rules and updated in the knowledge base, ensuring its timeliness and completeness. This allows the health status assessment model to continuously evolve with the accumulation of clinical practice. When encountering similar cases in the future, the model can use newly added rules to constrain its generation strategy, avoiding repeating past mistakes and achieving adaptive optimization of the health guidance model. This significantly improves the accuracy and professional consistency of personalized health recommendations.

[0038] It should be added that after the cerebrovascular disease risk assessment results and health guidance recommendations are sent to the target users, the target users' terminals display the sources of risk factors and the reasoning basis in the initial health guidance recommendations; in response to the professional's editing operation on the recommendation content, the text content of the health guidance recommendations is updated; the content before modification, the content after modification, and the type of modification operation are recorded, and correction log data is generated.

[0039] Specifically, after receiving the results returned by the target user, the process also includes: obtaining the user's case based on the results and the corresponding user's multi-source information; extracting and classifying the risk factors in the user's case when the user meets the criteria for stroke risk based on the user's case; updating the classified risk factors and corresponding user characteristics to the cerebrovascular disease knowledge base; and structuring the health guidance suggestions in the results, converting them into standardized intervention strategy items, and updating them to the cerebrovascular disease knowledge base.

[0040] It should be noted that by extracting and classifying risk factors from user cases that meet the criteria for high-risk stroke populations, different types of medium- and high-risk characteristics are continuously accumulated. This not only enriches the definition boundaries of high-risk stroke populations in the knowledge base but also strengthens the model's ability to identify specific combinations of risk factors. This allows the model to more accurately warn of high-risk risks when encountering users with similar characteristics in the future. Furthermore, the health guidance suggestions returned by target users are transformed into reusable knowledge assets. As the knowledge base in the field of cerebrovascular diseases is used over time, it can spontaneously absorb the latest clinical evidence and intervention experience. When encountering users with similar symptoms in the future, the model can directly call these validated standardized entries to quickly generate scientific and standardized health guidance, significantly improving service efficiency and quality.

[0041] In addition, based on the cerebrovascular disease risk assessment results and health guidance recommendations, the differences between these and the results returned by the target users are analyzed, and correction features reflecting the logic of professional knowledge are extracted. These include: semantically aligning the cerebrovascular disease risk assessment results and health guidance recommendations with the results returned by the target users to identify the target text fragments that have been modified; classifying the target text fragments to determine the modification type, which includes at least one of medical error correction, personalized supplementation, and expression optimization; and invoking the corresponding entity relationship extraction strategy according to the modification type to extract key entities and relationships between entities from the target text fragments, constructing condition-conclusion pairs as correction features reflecting the logic of professional knowledge.

[0042] It should be added that the entity relationship extraction strategy includes: when the modification type is determined to be medical error correction, extracting key entities and entity relationships containing medical contraindications or error risk factors, and combining them with the intervention measures after correction to construct conditional conclusion pairs; when the modification type is determined to be personalized supplementation, extracting key entities and entity relationships containing user-personalized attribute characteristics of non-general medical advice, and combining them with the newly added specific intervention plan to construct conditional conclusion pairs; when the modification type is determined to be expression optimization, extracting key entities and entity relationships containing the initial descriptive context, and combining them with text expression paradigms that conform to medical standards or user understanding habits to construct conditional conclusion pairs.

[0043] Furthermore, the modified features are transformed into structured knowledge memory rules, including: semantically abstracting the modified features to extract the medical entities involved in the modification and the modification intent; associating and matching the medical entities with user multi-source information to locate the user feature conditions that trigger the modification from the user multi-source information; determining the rule storage attributes according to the modification intent, establishing a mapping relationship between user feature conditions and the modified health guidance suggestions, forming a general guidance rule or preference pattern containing rule storage attributes, and obtaining knowledge memory rules. This allows the model to prioritize generating content that conforms to the modified suggestion conclusions when encountering similar users who meet the user feature conditions in the future, based on the rule storage attributes matching the corresponding generation strategy.

[0044] It should be added that medical entities are used to represent key information units with specific medical meanings; user characteristic conditions are used to represent the objective characteristics possessed by the user.

[0045] In summary, this invention, by acquiring multi-source information representing a user's comprehensive health status, enables the model to simultaneously consider multiple factors such as the user's physiological indicators, medical history, and lifestyle. This provides rich data support for generating in-depth assessment results. Furthermore, it incorporates medical knowledge fragments retrieved from a knowledge base in the field of cerebrovascular diseases to supplement the model with the implicit medical logic within the user's multi-source information. This effectively avoids the illusion problem that models are prone to in the medical field, ensuring that the assessment process is based on evidence and significantly improving the professionalism and credibility of the assessment results. The model not only outputs risk levels but also reveals deep pathological connections between different risk factors, elevating the assessment results from simple numerical predictions to in-depth diagnostic analysis with medical logic. This allows for the direct generation of customized health guidance suggestions based on the user's unique risk characteristics. While improving the accuracy of the assessment results, it ensures that the suggestions are not generic popular science but rather action guidelines tailored to the user's individual risk factor list, achieving integrated and precise output of risk and intervention, and a closed loop of personalized health management.

[0046] The following describes the cerebrovascular disease risk assessment device based on a large model intelligent agent provided by the present invention. The cerebrovascular disease risk assessment device based on a large model intelligent agent described below can be referred to in correspondence with the cerebrovascular disease risk assessment method based on a large model intelligent agent described above.

[0047] Figure 3 A device for assessing the risk of cerebrovascular diseases based on a large model intelligent agent is shown, characterized in that it includes: Information acquisition module 31 acquires multi-source information about the user, which represents the user's overall health status; The knowledge retrieval module 32 retrieves relevant medical guideline knowledge fragments based on multi-source user information and a knowledge base in the field of cerebrovascular diseases, serving as contextual reference material. The risk assessment module 33 inputs multi-source user information and contextual reference materials into the health status assessment model to obtain the health status assessment results output by the health status assessment model. The health status assessment results include cerebrovascular disease risk assessment results and health guidance suggestions. The health status assessment model is trained based on user multi-source training data and the corresponding health status assessment labels of the user multi-source training data.

[0048] It should be noted that the specific principles of the embodiments of the present invention are the same as those of the above method embodiments. For details, please refer to the method embodiments above. More detailed explanations will not be repeated here.

[0049] Figure 4 An example is a schematic diagram of the physical structure of an electronic device, such as... Figure 4 As shown, the electronic device may include a processor 410, a communications interface 420, a memory 430, and a communication bus 440. The processor 410, communications interface 420, and memory 430 communicate with each other via the communication bus 440. The processor 410 can call logical instructions in the memory 430 to execute a cerebrovascular disease risk assessment method based on a large-scale intelligent agent. This method includes: acquiring multi-source user information, which represents the user's comprehensive health status; acquiring corresponding medical guideline knowledge fragments based on the multi-source user information and a cerebrovascular disease domain knowledge base, as contextual reference material; inputting the multi-source user information and contextual reference material into a health status assessment model to obtain the health status assessment result output by the model, which includes a cerebrovascular disease risk assessment result and health guidance suggestions; wherein the health status assessment model is trained based on user multi-source training data and the corresponding health status assessment labels.

[0050] Furthermore, the logical instructions in the aforementioned memory 430 can be implemented as software functional units and, when sold or used as independent products, can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0051] On the other hand, the present invention also provides a computer program product, which includes a computer program that can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer can execute the cerebrovascular disease risk assessment method based on a large model intelligent agent provided by the above methods. The method includes: acquiring multi-source user information, which represents the user's comprehensive health status; acquiring corresponding medical guideline knowledge fragments based on the multi-source user information and a knowledge base in the field of cerebrovascular diseases, as contextual reference material; inputting the multi-source user information and contextual reference material into a health status assessment model to obtain a health status assessment result output by the health status assessment model, which includes a cerebrovascular disease risk assessment result and health guidance suggestions; wherein, the health status assessment model is trained based on user multi-source training data and the health status assessment labels corresponding to the user multi-source training data.

[0052] In another aspect, the present invention also provides a non-transitory computer-readable storage medium storing a computer program thereon. When executed by a processor, the computer program implements the cerebrovascular disease risk assessment method based on a large-scale intelligent agent provided by the above methods. The method includes: acquiring multi-source user information, which represents the user's comprehensive health status; acquiring corresponding medical guideline knowledge fragments based on the multi-source user information and a cerebrovascular disease domain knowledge base as contextual reference material; inputting the multi-source user information and contextual reference material into a health status assessment model to obtain a health status assessment result output by the health status assessment model, which includes a cerebrovascular disease risk assessment result and health guidance suggestions; wherein the health status assessment model is trained based on user multi-source training data and the health status assessment labels corresponding to the user multi-source training data.

[0053] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.

[0054] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.

[0055] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for assessing the risk of cerebrovascular diseases based on a large-scale intelligent agent model, characterized in that, include: Acquire multi-source user information, which represents the user's overall health status; Based on the multi-source user information and combined with the knowledge base in the field of cerebrovascular diseases, corresponding medical guideline knowledge fragments are obtained as contextual reference materials; The multi-source user information and the contextual reference material are input into the health status assessment model to obtain the health status assessment result output by the health status assessment model. The health status assessment result includes the cerebrovascular disease risk assessment result and health guidance suggestions. The health status assessment model is trained based on the user's multi-source training data and the health status assessment labels corresponding to the user's multi-source training data.

2. The method for assessing the risk of cerebrovascular diseases based on a large-scale intelligent agent according to claim 1, characterized in that, The user's multi-source information includes basic user information, past medical history and family history of genetic diseases, physical examination and test indicators, lifestyle information and subjective health description information; Based on the aforementioned multi-source user information and combined with a knowledge base in the field of cerebrovascular diseases, corresponding medical guideline knowledge fragments are obtained as contextual reference materials, including: The user basic information and the physical examination and examination indicators are subjected to structured encoding and numerical normalization to obtain structured feature vectors. Natural language processing and key entity extraction are performed on the past medical history and family genetic history, the lifestyle information and subjective health description information to obtain text semantic feature vectors. The structured feature vector and the text semantic feature vector are fused across modalities to obtain a multimodal fused feature vector. The multimodal fused feature vector is then used to retrieve corresponding medical guideline knowledge fragments from a pre-built knowledge base for cerebrovascular diseases, which serve as contextual reference material.

3. The method for assessing the risk of cerebrovascular diseases based on a large-scale intelligent agent according to claim 1, characterized in that, The health guidance recommendations include lifestyle adjustment suggestions, key intervention tips, and follow-up visits; The multi-source user information and the contextual reference material are input into the health status assessment model to obtain the health status assessment result output by the health status assessment model, including: The multi-source user information and the contextual reference material are input into the health status assessment model. Based on the contextual reference material, the multi-source user information is analyzed to identify the user's risk factors. Combined with the corresponding risk weights, a cerebrovascular disease risk assessment result containing a list of risk factors and a comprehensive risk level is generated. Based on the multi-source user information, identify the abnormal indicator features and key entities with the highest risk contribution, and generate key intervention prompts; Based on the list of risk factors, intervention targets are identified, corresponding intervention measures are matched, and corresponding lifestyle adjustment recommendations are generated. Based on the comprehensive risk level, the follow-up period is determined, and based on the abnormal indicator characteristics and key entities, the re-examination items are determined, and re-examination and follow-up recommendations are generated.

4. The method for assessing the risk of cerebrovascular diseases based on a large-scale intelligent agent according to claim 1, characterized in that, After obtaining the health status assessment results output by the health status assessment model, the following are included: The cerebrovascular disease risk assessment results and health guidance suggestions are sent to the target user, and the results returned by the target user are received; the results include the content after the target user confirms, modifies and / or supplements the cerebrovascular disease risk assessment results and health guidance suggestions; Based on the cerebrovascular disease risk assessment results and health guidance recommendations, the differences between these results and those returned by the target users are analyzed, and corrective features reflecting the logic of professional knowledge are extracted. The modified features are transformed into structured knowledge memory rules, and the knowledge base for the cerebrovascular disease domain is updated; the knowledge memory rules are used to constrain the generation strategy during model reasoning.

5. The method for assessing the risk of cerebrovascular diseases based on a large-scale intelligent agent according to claim 4, characterized in that, Based on the cerebrovascular disease risk assessment results and health guidance recommendations, the differences between these results and those returned by the target users are analyzed, and correction features reflecting the logic of professional knowledge are extracted, including: Semantically align the cerebrovascular disease risk assessment results and health guidance suggestions with the results returned by the target user to identify modified target text fragments. The target text fragment is classified to determine the modification type, which includes at least one of medical error correction, personalized supplementation, and expression optimization. Based on the modification type, the corresponding entity relationship extraction strategy is invoked to extract key entities and relationships between entities from the target text fragment, and conditional conclusion pairs are constructed as correction features reflecting professional knowledge logic.

6. The method for assessing the risk of cerebrovascular diseases based on a large-scale intelligent agent according to claim 5, characterized in that, The entity relationship extraction strategy includes: When the modification type is determined to be medical error correction, key entities containing medical contraindications or error risk factors and the relationships between entities are extracted, and conditional conclusion pairs are constructed by combining them with the intervention measures after correction. When the modification type is determined to be personalized supplement, the key entities and relationships between entities containing user personalized attribute features that include non-general medical advice are extracted, and combined with the newly added specific intervention plan, conditional conclusion is correct; When the modification type is determined to be expression optimization, key entities and relationships between entities containing the initial descriptive context are extracted, and conditional conclusion pairs are constructed by combining them with text expression paradigms that conform to medical standards or user understanding habits.

7. The method for assessing the risk of cerebrovascular diseases based on a large-scale intelligent agent according to claim 4, characterized in that, After receiving the result returned by the target user, the process also includes: Based on the results and the corresponding user multi-source information, the user's medical record is obtained; Based on the user's case, when it is determined that the user meets the criteria for a stroke-risk population, the risk factors in the user's case are extracted and classified. The classified risk factors and corresponding user characteristics are updated in the cerebrovascular disease knowledge base. The health guidance suggestions in the results are structured and transformed into standardized intervention strategy items, which are then updated in the cerebrovascular disease knowledge base.

8. The method for assessing the risk of cerebrovascular diseases based on a large-scale intelligent agent according to claim 4, characterized in that, The modified features are transformed into structured knowledge memory rules, including: The modified features are semantically abstracted to extract the medical entities involved in the modification and the modification intent. The medical entity is associated and matched with the user's multi-source information, and the user feature conditions that trigger the correction are located from the user's multi-source information. Based on the stated modification intent, rule storage attributes are determined, a mapping relationship is established between the user characteristic conditions and the revised health guidance suggestions, and a general guidance rule or preference pattern containing the rule storage attributes is formed, thus obtaining the knowledge memory rule.

9. The method for assessing the risk of cerebrovascular diseases based on a large-scale intelligent agent according to claim 1, characterized in that, Before inputting the multi-source user information into the health status assessment model, the following steps are included: Acquire human-computer collaborative interaction data during the historical health guidance generation process. The human-computer collaborative interaction data includes user multi-source training data and health status assessment labels corresponding to the user multi-source training data. The status assessment labels are obtained by correcting the initial assessment results and suggestions obtained by the large model based on the user multi-source training data. The user's multi-source training data is used as input data for training, and the risk assessment label and health guidance label are used as training labels to train the large model, thereby obtaining a health status assessment model for assessing the user's cerebrovascular disease risk and providing health guidance.

10. A cerebrovascular disease risk assessment device based on a large-scale intelligent agent model, characterized in that, include: The information acquisition module acquires multi-source information about the user, which represents the user's overall health status. The knowledge retrieval module retrieves relevant medical guideline knowledge fragments based on the multi-source user information and in conjunction with a knowledge base in the field of cerebrovascular diseases, as contextual reference material. The risk assessment module inputs the multi-source user information and the contextual reference materials into the health status assessment model to obtain the health status assessment results output by the health status assessment model. The health status assessment results include cerebrovascular disease risk assessment results and health guidance suggestions. The health status assessment model is trained based on user multi-source training data and the health status assessment labels corresponding to the user multi-source training data.

Images